From 5cebd7d735a71284c93ba81fc4a0fed8be916a17 Mon Sep 17 00:00:00 2001
From: qnsh <qinnianshun@gmail.com>
Date: Tue, 28 Oct 2025 14:58:32 +0800
Subject: [PATCH] =?UTF-8?q?ollama=20VLM=20=E6=94=AF=E6=8C=81=E5=A4=9A?=
 =?UTF-8?q?=E4=B8=AA=E5=9B=BE=E5=83=8F=E8=BE=93=E5=85=A5?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 locales/en/nodeDefs.json  |  6 +++---
 locales/zh/nodeDefs.json  |  4 ++--
 ollama/ollama_vlm_node.py | 34 +++++++++++++++++++++-------------
 3 files changed, 26 insertions(+), 18 deletions(-)

diff --git a/locales/en/nodeDefs.json b/locales/en/nodeDefs.json
index 582ee96..82adb64 100644
--- a/locales/en/nodeDefs.json
+++ b/locales/en/nodeDefs.json
@@ -57,9 +57,9 @@
         "display_name": "Ollama VLM API",
         "description": "This node uses the Ollama VLM model for image reasoning and analysis.",
         "inputs": {
-            "image":{
-                "name": "image",
-                "tooltip": "Image used for analysis"
+            "images":{
+                "name": "images",
+                "tooltip": "Images used for analysis"
             },
             "model": {
                 "name": "model"
diff --git a/locales/zh/nodeDefs.json b/locales/zh/nodeDefs.json
index 4f91bbe..5de2a05 100644
--- a/locales/zh/nodeDefs.json
+++ b/locales/zh/nodeDefs.json
@@ -58,9 +58,9 @@
         "display_name": "Ollama 视觉 API",
         "description": "这个节点使用Ollama VLM 模型进行图片推理分析",
         "inputs": {
-            "image":{
+            "images":{
                 "name": "图像",
-                "tooltip": "用于分析的图像"
+                "tooltip": "用于分析的图像(支持多个)"
             },
             "model": {
                 "name": "模型"
diff --git a/ollama/ollama_vlm_node.py b/ollama/ollama_vlm_node.py
index 81c6d8f..628cdcd 100644
--- a/ollama/ollama_vlm_node.py
+++ b/ollama/ollama_vlm_node.py
@@ -66,7 +66,7 @@ class OllamaVLM(io.ComfyNode):
             category="YCYY/API/text",
             inputs=[
                 io.Image.Input(
-                    "image",
+                    "images",
                     tooltip="Image used for analysis"
                 ),
                 io.String.Input(
@@ -106,7 +106,7 @@ class OllamaVLM(io.ComfyNode):
     #         return []
     # 执行 GeminiImage 节点
     @classmethod
-    def execute(cls,image, system_prompt, user_prompt, model) -> io.NodeOutput:
+    def execute(cls,images, system_prompt, user_prompt, model) -> io.NodeOutput:
         if not user_prompt:
             raise ValueError("User prompt cannot be empty")
 
@@ -123,19 +123,27 @@ class OllamaVLM(io.ComfyNode):
                 "content": system_prompt
             }
             payload["messages"].append(system_message)
-        image_base64 = tensor_to_base64_string(image)
-        user_message  ={
-            "role": "user",
-            "content": [
-                {
-                    "type": "text",
-                    "text": user_prompt
-                },
-                {
+
+        # 构建用户消息内容，支持多个图片
+        content = [
+            {
+                "type": "text",
+                "text": user_prompt
+            }
+        ]
+
+        # 处理多个图片
+        if images is not None:
+            for image_index in range(images.shape[0]):
+                image_base64 = tensor_to_base64_string(images[image_index].unsqueeze(0))
+                content.append({
                     "type": "image_url",
                     "image_url": f"data:image/png;base64,{image_base64}"
-                }
-            ]
+                })
+
+        user_message = {
+            "role": "user",
+            "content": content
         }
         payload["messages"].append(user_message)
         try: